Abstract: In recent years, short-term extreme wind fields prediction has become a research hotspot and academic frontier in the area of international wind engineering due to its vital role in structural safety. Accurate prediction of in-situ wind speed before the arrival of extreme wind fields is of great significance for the early warning of engineering structures safety and emergency protection. The traditional numerical weather prediction method is effective for extreme wind field prediction. However, due to insufficient spatial resolution and high consumption of computing resources, it is difficult to provide a real-time prediction of in-situ wind speed for engineering structures. With the rapid development of artificial intelligence technology, machine learning offers new ideas for solving the problems mentioned above. It is increasingly widely applied in short-term extreme wind fields prediction, showing broad application prospects. In this regard, this paper provides a comprehensive review of recent progress in the short-term extreme wind fields prediction using machine learning-based approaches. Firstly, the application principles and characteristics of time series models, machine learning models, and hybrid models in wind field prediction are reviewed. Subsequently, we classify and evaluate prevalent methods for short-term extreme wind field prediction, focusing on three predominant wind types: regular strong winds, typhoons, and thunderstorm winds. Their advantages and limitations are summarized. Finally, considering current research gaps and challenges in short-term prediction of extreme wind fields, potential future directions are proposed.